The rise of AI chatbots in customer service has sparked discussions about the potential job displacement of customer agents. However, industry experts agree that AI, at its current stage, is not ready to fully replace humans in customer-facing roles.
Recent incidents, such as a software engineer manipulating a chatbot to sell him a car for just a dollar, highlight the limitations and challenges of generative AI. One major issue is the phenomenon of persistent hallucinations, where AI models produce incorrect or nonsensical answers. While models like GPT-4 are being fine-tuned and trained with enterprise data, the risk of providing misleading information still exists.
Despite these challenges, generative AI can significantly enhance efficiency and customer experience in customer service. According to Sanjeev Menon, co-founder and head of product & tech at E42.ai, careful design and fine-tuning with specific data can elevate the performance of AI chatbots. However, it is crucial to fully understand the capabilities and limitations of these models.
Although many enterprises have integrated AI-powered chatbots into their platforms, this does not eliminate the need for human intervention. Human agents play a vital role in ensuring a positive and secure customer service interaction. Checks on prompt toxicity, data updates, and supervision during complex or sensitive situations are necessary to maintain accuracy and provide a reliable experience.
Gaurav Singh, founder and CEO at Verloop.io, suggests empowering human agents as gatekeepers to ensure quality control. While the majority of queries can be effectively handled by AI chatbots, uncertain situations require seamless transfer to human agents for verification and editing.
Small Language Models (SLMs) offer a more tailored solution to industry-specific customer service needs. These models can be fine-tuned or trained specifically for a particular domain, allowing for better understanding of industry-specific terminology and context. Enterprises have more control over the training process and can align the model with their specific requirements.
Nonetheless, the risk of hallucination still exists in SLMs. Yellow.ai, for example, uses a maker-checker model setup to validate the relevance and accuracy of responses. Despite these advancements, a domain-specific model with human involvement remains the best approach to mitigate risks and provide excellent customer service.
In conclusion, while generative AI has its benefits, a hybrid approach that combines the strengths of AI and human agents is crucial for delivering efficient and reliable customer service. Striking a balance between AI automation and human intervention ensures that emotional intelligence, nuanced understanding, and complex problem-solving are not compromised.